Multiobjective blood pump impeller optimization with three response surface methods and prototype stator casting for an integrated motor pump

Cardiovascular diseases are a leading cause of mortality worldwide; rotary blood pumps are a key therapy. Prior optimization studies often treated few geometric variables with single-objective formulations. We address multi-conflicting objectives under a comprehensive parameterization: fifteen blade and flow-path design variables, sampled by a Design of Experiments plan generating 290 Computational fluid dynamics cases validated against the Food and Drug Administration benchmark pump. Kriging, Genetic Aggregation, and Neural Networks were trained as response-surface surrogates, enabling a head-to-head comparison of Response Surface Method techniques. A Multi-Objective Genetic Algorithm simultaneously maximized outlet pressure and minimized mean scalar shear stress. Three Pareto-optimal designs were examined for global and local sensitivities. The Neural-Network surrogate provided the best fidelity for optimization; the most-optimized design achieved a 23% increase in outlet pressure and a 26% reduction in mSS relative to the baseline. To bridge simulation and hardware, an Fe₃O₄ cast stator was also rapidly prototyped. These results demonstrate a complete pipeline from Design of Experiments based surrogate modeling and multiobjective optimization to prototype realization for integrated blood pump drives.